WP4 Tools - Farm Setup

WP4 task is the setup and management of farms, the so called fabric grid component. Fabric elements are:

• installation and configuration
• monitoring of sanity and fault tolerance
• configuration management and update
• resource management, like farm status, queue management, etc.

At present farm setup is handled by LCFG a client-server tool that performs node clustering. In any site with grid resources, even minimal as a single UI, there should be a node acting as LCFG server. The server stores all tools required to install and maintain nodes of the farming. Each node is installed from the server via a startup image (at present stored on floppy, in next releases directly from the network)

The startup image boots the node and starts Linux installation from the LCFG server. The procedure mimics the standard installation but with a more granular configuration. For each node the LCFG server must store a configuration table so that the node is installed as CE, SE, UI, WN as planned in the local site. All components must be installed from rpms as declared in the configuration file. LCFG is much more than an installation tool as all operations on the nodes are controlled and started from the LCFG sever. Configurable and variable items are called objects that are defined and maintained on the LCFG server for each node. Just to make an easy example, the user accounting is an object and new users are created on each node from the LCFG server.

In the same way sofware pakages are maintained on the server and cron jobs may perform automatic updates at conveninet time intervals. A very important feature of LCFG is the cleanup of non-LCFG actions at node reboot, therefore any user added on a node using useradd or any software pack added manually will be erased when the node reboots.

Present LCFG limitations are:

1. a complex description language for objects (C define syntax)
2. Linux extended partitions are not supported
3. LCFG server and nodes must be installed from scratch, this is a requirement to avoid unresolved or conflicting rpm dependencies

Future foreseen improvements are:

1. the implementation of a high level configuration description language (HLD)
2. implementation of a configuration database (CDB)
3. implementation of a node maintainance GUI
4. implementation of LCFG light to allow LCFG installation on working nodes keeping the existing operating system